The longleaf pine (Pinus palustris) savanna ecosystem is an imperiled, fire-dominated community that supports exceptionally high levels of species richness and endemism. Area of this community has declined by more than 95% due to unsustainable logging, fire suppression, and changes in land-use practices. In recent decades, efforts to restore fire-dominated communities like longleaf pine savanna have gained popularity, especially in light of benefits to charismatic species like the northern bobwhite (Colinus virginianus). Although reptiles are important members of this ecological community, far less information exists as to how this group responds to longleaf pine management, especially when game bird conservation is a primary management focus. Although bobwhite management in these systems is mostly synonymous with longleaf pine restoration, additional conservation practices aimed at game birds (e.g., promoting fallow fields, supplemental feeding, meso-carnivore control, cross sectional mowing, etc.) might affect the extent to which squamates benefit from habitat management. To better understand how squamate reptiles may benefit from longleaf pine savanna managed for northern bobwhites, we surveyed for six-lined racerunners (Aspidoscelis sexlineata) across a large, contiguous tract of longleaf pine with varied land cover characteristics, managed to maximize the conservation of northern bobwhites. Racerunner detection probability on transect surveys was low ( = 0.23) however, occupancy probability, was relatively high ( = 0.60) across the property and driven by percent open ground (positive; 25m scale), percent grass cover (negative; 25m scale), and percent wetland (negative; 100m scale). Our findings support those of past studies about six-lined racerunners in longleaf pine savannas suggesting the species thrives in the context of a short fire rotation (e.g., 2-3 years), even when game bird management is a primary objective of conservation efforts. Racerunners may also specialize on microhabitats (e.g., upland areas with relatively high bare ground cover) that occur most frequently in recently burned portions of bobwhite management units.

Study area and sampling locations. We studied six-lined racerunner ecology in Brunswick County, North Carolina. The study area falls into the southeastern Coastal Plain which is an area characterized by a subtropical climate, and nutrient-poor, well-drained soils (Peet and Allard 1993). Within Brunswick County, we surveyed a 1,850 ha contiguous stand of longleaf pine savanna managed for hunted game species, especially northern bobwhites. Although the dominant land cover was mature longleaf pine savanna, there were other cover types naturally interspersed: wetlands, early successional longleaf pine, dirt roads/firebreaks, and open water (ponds, bays, etc.). There were also fallow fields throughout the property (usually 1-2 ha in size) that were sown with partridge-pea (Chamaecrista fasciculata) and/or ragweed (Ambrosia artemisiifolia) to support northern bobwhite brood-rearing. In addition to fallow field cultivation, the property implemented other conservation practices to support quail: supplemental feeding (~3 bushels/ac/year of wheat and sorghum), year-round mesocarnivore trapping, and prescribed fire with a rotation cycle of every 2 years (half the property burned each year).

Six-lined racerunner surveys. To establish sampling locations, we gridded the study area into 100 cells with a sampling point plotted in the center of each. This approach generated sampling locations that were each ~400m from its nearest neighbor. We used this systematic sampling approach to maximize coverage of the study area while maintaining the spatial independence of survey locations. Each of these 100 sampling locations became the centroid for a visual transect for six-lined racerunners. We conducted repeat surveys along 50m transects, oriented south to north, centered over the survey points, from 3 July through 22 August, 2023, between 0630 and 2300. We visited each of the 100 points three times in random order (100 sites x 3 visits = 300 total surveys). We did not restrict surveys to particular weather conditions within this window because we were interested in understanding the factors that might explain variation in detection probability. Prior to each survey, we recorded survey covariates: cloud cover (estimated to the nearest 25%), precipitation (none, mist, light rain, heavy rain), temperature, Beaufort Wind Index, ordinal date, and time of day. We walked each transect for two minutes (i.e., 25 meters per minute) and recorded whether six-lined racerunners were detected during each survey (1/0). Because we visited each site three times, we generated a detection history of 1s and 0s that served as our response data for occupancy models (MacKenzie et al. 2002), described below.

Vegetation surveys. We sampled local vegetation (within 25m) of each transect (hereafter, “microhabitat”) once between 30 July and 12 August. We estimated percent cover of vegetation strata and vegetation within 25m of each site using the same transect as surveyed for racerunners, sampled with an ocular tube (James and Shugart 1970) and a Nudds board (Nudds 1977), respectively. We estimated percent cover using an ocular tube at 10 “stops” per site, one at each of the following distances along the transect: 5m, 10m, 15m, 20m, and 25m, for north and south. At each “stop”, we recorded: bare ground, leaf litter, coarse woody debris, moss, grass, forb, shrub, fern, sapling, canopy. Bare ground was defined as un-vegetated soil. Leaf litter was defined as any organic debris smaller than 10cm. Coarse woody debris was any organic debris larger than 10cm in diameter. We considered moss to be any live, non-vascular plant. Grasses were any graminoid. Forbs were any broad-leafed dicotyledon. We defined a shrub as a woody plant with stems branching above ground. Ferns were seedless vascular plants. Saplings were any tree less than 10cm in diameter-at-breast-height (DBH); trees larger than 10cm in DBH were considered canopy trees (McNeil et al. 2018). Finally, we assessed vegetation density via a Nudds Board (2m tall with 20 squares, each 20x20cm in size), placed at point center. We read the Nudds board from 10 meters north and south of the board from a height of 1m. Each reading consisted of a visual assessment of the number of squares (out of 20 possible) obscured at least halfway by vegetation. We averaged the two Nudds board readings at each site and the number of “hits” for each vegetation stratum (/10 stops) for percent cover of each microhabitat stratum.

Remote sensed data. To assess habitat at a broader spatial extent, we extracted metrics from a land cover map (unpl. data) developed in qGIS (QGIS.org). Relevant land cover values included: 1. mature pine, 2. open grassland (including grassy areas with short, sapling pines), 3. wetland, 4. brood field, and 5. open water. We generated rasters that depicted ‘distance to nearest…’ for each of these covariates with the raster package in R (Hijmans 2023). We also generated distance to nearest road. In addition to these ‘distance to nearest…’ variables, we calculated the “percent cover within 100m” for each variable (except ‘road’ because it was a line vector file). Finally, we extracted the normalized difference vegetation index (NDVI) at each point using a raster developed for the property in 2019, obtained from the property manager. 

Analyses. We created single-season occupancy models in the R package unmarked (MacKenzie et al. 2002, Fiske and Chandler 2011, R Core Team 2023) to assess factors influencing six-lined racerunner occupancy. We only specified univariate models in this study (no additive covariates) to avoid over-parameterizing our models. We began by fitting a ‘detection’ model set where we held occupancy probability constant and specified models with each of the following survey covariates: temperature at the time of survey, ordinal date, Beaufort Wind Index, minutes since sunrise, cloud cover, and two habitat covariates that might explain observers’ ability to see racerunners: NDVI and vegetation density. We also included quadratic terms for temperature and minutes since sunrise because these variables might exhibit a nonlinear trend. We also included a null (intercept only) detection model for comparison: p(.),ᴪ(.) in all model sets. We ranked all models in descending order of Akaike’s Information Criterion adjusted for small sample size (AIC_c; Akaike 1973, Burnham and Anderson 2002). We also assessed the biological importance of covariates by examining whether β 95% confidence intervals overlapped zero, interpreting those that did so as having weak effects on detection/occupancy (Arnold 2010). After identifying a top detection model, we incorporated its detection terms into all of our following occupancy models. This second and final model set included univariate models for each of the microhabitat variables and each of the landscape variables. Finally, we also assessed model goodness-of-fit by calculating  via the MacKenzie and Bailey goodness-of-fit test (MacKenzie and Bailey 2004, Kéry and Royle 2015). If our top model was overdispersed ( > 1.0), we adjusted our model ranking using Quasi AIC_c (QAIC_c).